A US judge has dismissed some of the claims made by writers in a copyright infringement lawsuit against OpenAI, though gave the wordsmiths another chance to amend their complaint.

The case - Paul Tremblay et al vs OpenAI - kicked off in 2023 when novelists Paul Tremblay, Christopher Golden, and Richard Kadrey, and writer-comedian-actress Sarah Silverman accused OpenAI of illegally scraping their work without consent to train the AI champion's large language models.

The creators claimed that ChatGPT produced accurate summaries of their books and offered that as evidence that their writing had been ripped off. Since OpenAI's neural networks learn to generate text from its training data, the group argued that its output should be considered a "derivative work" of their IP.

The plaintiffs also alleged that OpenAI’s model deliberately omitted so-called copyright management information, or CMI - think books' ISBN numbers and authors' names - when it produced output based on their works. They also accused the startup of unfair competition, negligence, and unjust enrichment.

All in all, the writers are upset that, as alleged, OpenAI not only used copyrighted work without permission and recompense to train its models, its model generates prose that closely apes their own, which one might say would hinder their ability to profit from that work.

Federal district Judge Araceli Martínez-Olguín, sitting in northern California, was asked by OpenAI to dismiss the authors' claims in August.

In a fresh order [PDF] released on Monday, Martínez-Olguín delivered the bad news for the scribes.

"Plaintiffs fail to explain what the outputs entail or allege that any particular output is substantially similar - or similar at all - to their books. Accordingly, the court dismisses the vicarious copyright infringement claim," she wrote. She also opined that the authors couldn't prove that CMI had been stripped from the training data or that its absence indicated an intent to hide any copyright infringement.

Claims of unlawful business practices, fraudulent conduct, negligence, and unjust enrichment were similarly dismissed.

The judge did allow a claim of unfair business practices to proceed.

"Assuming the truth of plaintiffs' allegations - that defendants used plaintiffs' copyrighted works to train their language models for commercial profit - the court concludes that defendants' conduct may constitute an unfair practice," Martínez-Olguín wrote.

Although this case against OpenAI has been narrowed, it clearly isn't over yet. The plaintiffs have been given another opportunity to amend their initial arguments alleging violation of copyright by filing a fresh complaint before March 13.

The Register has asked OpenAI and a lawyer representing the plaintiffs for comment. We'll let you know if they have anything worth saying. ®

https://www.theregister.com//2024/02/14/tremblay_vs_openai_update/